I am reposting my message from April 8th because I never received a response to
the original post:
Dear R-listers,
I am an MD and clinical epidemiologist developing a measure of comorbidity
severity for patients with liver disease. Having developed my comorbidity score
as the linear predictor from a Cox regression model I want to compare the
discriminative ability of my comorbidity measure with the "old" comorbidity
measure, Charlson's Comorbidity Index. I have nearly 10,000 deaths and 36
candidate comorbidities.
I wish to compare the discrimination of the two comorbidity measures, i.e. I
have two non-nested Cox models. I get the following output with
> rcorrp.cens(myscore.lp, charlson.lp, Surv(time, dead), method=1):
x1 = My comorbidity score, x2 = Charlson
[,1]
Dxy "-0.0605"
S.D. "0.00648"
x1 more concordant "0.4697"
x2 more concordant "0.5302"
n "1.369e+04"
missing "0"
uncensored "9411"
Relevant Pairs "1.587e+08"
Uncertain "2.861e+07"
C X1 "0.395"
C X2 "0.401"
Dxy X1 "-0.21"
Dxy X2 "-0.198"
I am aware that because a high hazard means short survival I must subtract C X1
and C X2 from 1, so my comorbidity score has marginally better discrimination
than the Charlson score (C = 0.605 vs. 0.599). Question: Is it true that my
score is more discriminative than the Charlson score in 53% of patient pairs?
I have done the same analysis with 'method = 2', i.e.
> rcorrp.cens(myscore.lp, charlson.lp, Surv(time, dead), method=2):
x1 = My comorbidity score, x2 = Charlson
[,1]
Dxy "-0.006002"
S.D. "0.001102"
x1 more concordant "0.04018"
x2 more concordant "0.04618"
n "1.369e+04"
missing "0"
uncensored "9411"
Relevant Pairs "1.587e+08"
Uncertain "2.861e+07"
C X1 "0.395"
C X2 "0.401"
Dxy X1 "-0.21"
Dxy X2 "-0.198"
Question: How do I interpret the 'x1/x2 more concordant' numbers in a Cox
regression setting? My guess: My comorbidity score concordant in 4.6% of pairs
in which Charlson's score is not. And Charlson's score is concordant in 4.0% of
pairs in which my comorbidity score is not.
Thank you in advance for your insight and help.
Best regards,
Peter Jepsen
Aarhus, Denmark
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
